As robots move beyond rigid, rule-based behavior, a new class of intelligent agents is emerging—robots that take initiative, act with purpose, and collaborate with humans. This talk explores the shift toward agentic AI, highlighting key design patterns and recent breakthroughs that have the potential to make autonomy not just possible, but useful. From language model-based robots that offer support only when truly needed, to systems that plan for more long-term tasks using extensible tool libraries, agentic AI is reshaping how robots perceive, decide, and interact. Drawing on real-world prototypes and research at the Honda Research Institute, this talk gives an overview of how robots can detect needs, choose when to act, and explain their behavior—(...)

- part of Session 1 "Autonomy and Interaction in Robotics" -